docs(vk): RX_VK_READBACK_LATE=3980,40,2000 on a discrete RADV card (RX 9070 XT) #1668

Open
Molier wants to merge 1 commit from Molier/Longshot:docs/readback-late-radv into main
Contributor

Summary

Longshot already has RX_VK_READBACK_LATE (#238, kept by #1763's renderer-threads rewrite: gpu/vk/switches.def
row and vk_replay.cpp's readback_late_wanted are unchanged on main @ 89c3d21e — verified by reading both): a
rule names GPU-result readbacks read without their write-back wait, off by default except the culling camera block
(0x3980 bytes, read without its wait on every PC) and, on handhelds, the newest submission's results too. This PR
only adds a second, independent measurement: on a discrete RADV card (RX 9070 XT), RX_VK_LOG_READBACK_RANGES=1
found the same dominant non-meter range (0x3980) the Steam Deck OLED PR found, plus two further sizes worth a rule
on this card (0x40, 0x2000); the exposure meter's own ranges are correctly left out by the switch's existing rule.
No code change — the switch, its rule syntax and the logging already do everything this needs. The text now lives in
docs/dev/reference/threads.md (#1802 moved docs/development.md there) plus a new docs/tested/ entry.

Still true after #1763: verified by reading gpu/vk/vk_replay.cpp:3407 (readback_late_wanted) and
gpu/vk/switches.def's RX_VK_READBACK_LATE row on upstream/main @ 89c3d21e — the default behaviour this PR's
numbers were measured against has not changed.

Testing

Tested: no code changed; this is a documentation-only update. No test in the current suite validates
docs/tested/'s format (tests/test_tested_configurations.py, named in the original commit, does not exist on
main @ 89c3d21e or in this rebase — the file this PR adds, docs/tested/2026-11-16-readback-late-oscar-run01.md,
follows the existing entries' table format in that folder by eye).

Still to test: nothing further; this documents a measurement already taken.

Measurements

Verdict: the measured rule (RX_VK_READBACK_LATE=3980,40,2000) helped on a discrete RADV card, on top of the
existing default.

Setup: Linux, headless, RX 9070 XT 16 GB (RADV/Mesa), Ryzen 7 5800X; 2560x1440, render resolution fixed 100%, frame
generation off, uncapped; a heavy six-leg route (Market/Castle Town/Hyrule Field x2) and a moving Hyrule Field route,
3 interleaved runs an arm, median frame time.

Before: main, RX_VK_READBACK_LATE unset After: RX_VK_READBACK_LATE=3980,40,2000
heavy c1_market — -7.9%
heavy c1_castle — -10.7%
heavy c2_market — -9.1%
heavy c2_castle — -7.8%
heavy c2_field — -3.6%
Hyrule Field run_a — -8.1%

(c1_field, run_b and load fell inside the 3-run noise band and are not claimed.) No exposure pulse found by a direct
visual compare; a luma check across the pairs could not separate a small exposure shift from camera jitter from one
run each, so that part is not established, not ruled out (noted in docs/tested/).

Risks

None: no code change, a documentation addition naming a rule a player or tester can already set today. The luma
caveat above is written into the docs/tested/ entry so a reader does not take "no exposure pulse" as fully proven.

Checklist

Ready for review

  • One topic, on current main; the title is a Conventional Commit.
  • Tested on my own PC, as Testing says (a documented measurement, not a code change to re-run).
  • No test regresses (none governs this file on main).
  • Measured against main (see Measurements; the switch's default is unchanged by #1763).
  • No game files, build output or hand-edited generated files; third_party/ untouched.

When it applies

  • Docs updated (this PR is the doc update; moved to docs/dev/reference/threads.md for #1802).
  • AI assistance recorded with Assisted-by: trailers.

Ready to merge

  • Approved in review, every comment answered.
  • CI passes on the head.
## Summary Longshot already has `RX_VK_READBACK_LATE` (#238, kept by #1763's renderer-threads rewrite: `gpu/vk/switches.def` row and `vk_replay.cpp`'s `readback_late_wanted` are unchanged on `main @ 89c3d21e` — verified by reading both): a rule names GPU-result readbacks read without their write-back wait, off by default except the culling camera block (0x3980 bytes, read without its wait on every PC) and, on handhelds, the newest submission's results too. This PR only adds a second, independent measurement: on a discrete RADV card (RX 9070 XT), `RX_VK_LOG_READBACK_RANGES=1` found the same dominant non-meter range (0x3980) the Steam Deck OLED PR found, plus two further sizes worth a rule on this card (0x40, 0x2000); the exposure meter's own ranges are correctly left out by the switch's existing rule. No code change — the switch, its rule syntax and the logging already do everything this needs. The text now lives in `docs/dev/reference/threads.md` (#1802 moved `docs/development.md` there) plus a new `docs/tested/` entry. **Still true after #1763:** verified by reading `gpu/vk/vk_replay.cpp:3407` (`readback_late_wanted`) and `gpu/vk/switches.def`'s `RX_VK_READBACK_LATE` row on `upstream/main @ 89c3d21e` — the default behaviour this PR's numbers were measured against has not changed. ## Testing **Tested:** no code changed; this is a documentation-only update. No test in the current suite validates `docs/tested/`'s format (`tests/test_tested_configurations.py`, named in the original commit, does not exist on `main @ 89c3d21e` or in this rebase — the file this PR adds, `docs/tested/2026-11-16-readback-late-oscar-run01.md`, follows the existing entries' table format in that folder by eye). **Still to test:** nothing further; this documents a measurement already taken. ## Measurements **Verdict:** the measured rule (`RX_VK_READBACK_LATE=3980,40,2000`) helped on a discrete RADV card, on top of the existing default. Setup: Linux, headless, RX 9070 XT 16 GB (RADV/Mesa), Ryzen 7 5800X; 2560x1440, render resolution fixed 100%, frame generation off, uncapped; a heavy six-leg route (Market/Castle Town/Hyrule Field x2) and a moving Hyrule Field route, 3 interleaved runs an arm, median frame time. | | Before: main, RX_VK_READBACK_LATE unset | After: RX_VK_READBACK_LATE=3980,40,2000 | |---|---|---| | heavy c1_market | — | -7.9% | | heavy c1_castle | — | -10.7% | | heavy c2_market | — | -9.1% | | heavy c2_castle | — | -7.8% | | heavy c2_field | — | -3.6% | | Hyrule Field run_a | — | -8.1% | (c1_field, run_b and load fell inside the 3-run noise band and are not claimed.) No exposure pulse found by a direct visual compare; a luma check across the pairs could not separate a small exposure shift from camera jitter from one run each, so that part is not established, not ruled out (noted in `docs/tested/`). ## Risks None: no code change, a documentation addition naming a rule a player or tester can already set today. The luma caveat above is written into the `docs/tested/` entry so a reader does not take "no exposure pulse" as fully proven. ## Checklist ### Ready for review - [x] One topic, on current `main`; the title is a Conventional Commit. - [x] Tested on my own PC, as Testing says (a documented measurement, not a code change to re-run). - [x] No test regresses (none governs this file on `main`). - [x] Measured against `main` (see Measurements; the switch's default is unchanged by #1763). - [x] No game files, build output or hand-edited generated files; `third_party/` untouched. ### When it applies - [x] Docs updated (this PR is the doc update; moved to `docs/dev/reference/threads.md` for #1802). - [x] AI assistance recorded with `Assisted-by:` trailers. ### Ready to merge - [ ] Approved in review, every comment answered. - [ ] CI passes on the head.
Molier force-pushed docs/readback-late-radv from be03f400aa
Some checks failed
Website / Website on Cloudflare (pull_request) Blocked by required conditions
PR description / Description follows the template (pull_request_target) Failing after 6s
Website / Website on Cloudflare (pull_request_target) Has been skipped
CI / Conventional Commit title (pull_request) Has been cancelled
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been cancelled
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been cancelled
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been cancelled
CI / Workflow lint (actionlint) (pull_request) Has been cancelled
Guide screenshots / Guide images and local UI source review (pull_request) Has been cancelled
Third-party parts / As their notes say (pull_request) Has been cancelled
Third-party parts / The same as upstream (pull_request) Has been cancelled
to 33dbda90b2
Some checks failed
PR description / Description follows the template (pull_request_target) Failing after 7s
CI / Conventional Commit title (pull_request) Has been cancelled
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been cancelled
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been cancelled
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been cancelled
CI / Workflow lint (actionlint) (pull_request) Has been cancelled
Guide screenshots / Guide images and local UI source review (pull_request) Has been cancelled
2026-11-18 23:46:51 +00:00
Compare
Molier force-pushed docs/readback-late-radv from 33dbda90b2
Some checks failed
PR description / Description follows the template (pull_request_target) Failing after 7s
CI / Conventional Commit title (pull_request) Has been cancelled
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been cancelled
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been cancelled
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been cancelled
CI / Workflow lint (actionlint) (pull_request) Has been cancelled
Guide screenshots / Guide images and local UI source review (pull_request) Has been cancelled
to 14eac40d86
Some checks failed
CI / Conventional Commit title (pull_request) Blocked by required conditions
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Blocked by required conditions
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Blocked by required conditions
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Blocked by required conditions
CI / Workflow lint (actionlint) (pull_request) Blocked by required conditions
CI / Website build and checks (pull_request) Blocked by required conditions
Guide screenshots / Guide images and local UI source review (pull_request) Blocked by required conditions
Third-party parts / As their notes say (pull_request) Blocked by required conditions
Third-party parts / The same as upstream (pull_request) Blocked by required conditions
Website / Website on Cloudflare (pull_request) Blocked by required conditions
PR description / Description follows the template (pull_request_target) Failing after 14s
2026-11-20 14:38:11 +00:00
Compare
Author
Contributor

Rebased onto main @ 89c3d21e. RX_VK_READBACK_LATE's default and readback_late_wanted are unchanged by #1763, so the measurement this documents still applies. The paragraph it extends moved to docs/dev/reference/threads.md in #1802; the added sentence moved with it.

Rebased onto `main` @ `89c3d21e`. `RX_VK_READBACK_LATE`'s default and `readback_late_wanted` are unchanged by #1763, so the measurement this documents still applies. The paragraph it extends moved to `docs/dev/reference/threads.md` in #1802; the added sentence moved with it.
Some checks failed
CI / Conventional Commit title (pull_request) Blocked by required conditions
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Blocked by required conditions
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Blocked by required conditions
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Blocked by required conditions
CI / Workflow lint (actionlint) (pull_request) Blocked by required conditions
CI / Website build and checks (pull_request) Blocked by required conditions
Guide screenshots / Guide images and local UI source review (pull_request) Blocked by required conditions
Third-party parts / As their notes say (pull_request) Blocked by required conditions
Third-party parts / The same as upstream (pull_request) Blocked by required conditions
Website / Website on Cloudflare (pull_request) Blocked by required conditions
PR description / Description follows the template (pull_request_target) Failing after 14s
This pull request doesn't have enough approvals yet. 0 of 1 approvals granted.
Some workflows are waiting to be reviewed.
This branch is out-of-date with the base branch
You are not authorized to merge this pull request.
View command line instructions

Checkout

From your project repository, check out a new branch and test the changes.
git fetch -u 'docs/readback-late-radv:Molier-docs/readback-late-radv'
git switch 'Molier-docs/readback-late-radv'
Sign in to join this conversation.
No reviewers
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
MagicBeanDev/Longshot!1668
No description provided.